Analysis of the MDF production process by means of statistical process modelling

The production of wood-based panels is a complex process which is influenced by various inter-relations between process parameters, such as pressure, temperature, moisture content. The relationships between process parameters and panel quality characteristics could not be fully described yet. One method of describing such relationships is the statistical process modeling. The goal of the project presented was to develop process models for the production of MDF and to evaluate the quality of the models with regard to prediction of panel properties. The object of the study was the production of 7.8 and 19 mm thick MDF panels on a continuously operating forming and pressing line. Most of the data were obtained from inline data loggers. Some additional variables, such as fibre length, were recorded and incorporated in the data sets by hand. 49 data sets were recorded to develop process models for the 7.8 mm panels, 43 data sets were recorded for the 19 mm panels (observation range). Samples of the panels studied were tested for density, internal bond and bending strength (MOR) as well as thickness swelling during 2 and 24 hours soaking in water, respectively. Univariate, linear regression analysis was employed to develop the process models. Process variables were selected by two statistical and two technology-based selection methods to form the model function. The evaluation of the models was performed on the basis of 10 additional data sets each, for 7.8 and 19 mm panels (prediction range). The model-based predictions of panel properties were compared with the actual test values of the prediction range and the standard deviation within the observation range. Modelling results differ considerably for the two panel types and the different panel properties investigated with regard to the number of variables in the models and the prediction accuracy. Models of satisfactory prediction quality could be developed on the basis of all four selection methods for process variables while the combination of methods did not lead to meaningful selections of variables. Statistical process models based on extensive data collections are valuable tools for sensitivity analyses and optimisation of normal MDF production processes. The identification of weak points in the production process, i. e. extensively varying process variables, must primarily lead to technical solutions of the respective problem and also to a continuous visualisation and control of the relevant process data. Furthermore, the collection of data needed for the process model already allows detailed process control and documentation as well as the integrated control of all panel production steps.